Open Access
Table 1.
Details on the neural network architecture and hyper-parameters used in this work.
| Layer type | Layer | Activation | Neurons |
|---|---|---|---|
| Encoder | Input | – | 48 |
| Layer 1 | Mish | 48 | |
| Layer 2 | Mish | 48 | |
| Output | – | 48 | |
| Baycenter representation in Λ | |||
| Decoder | Input | – | 48 |
| Layer 1 | Mish | 48 | |
| Layer 2 | Mish | 48 | |
| Output | – | 48 | |
| Parameters | Values | ||
| Optimiser | Adam | ||
| Step size | 10−3 | ||
| Batch size | 32 | ||
| Batch normalisation | Yes | ||
| Iterations | 25 000 | ||
| Residual parameter (ε0) | 0.1 | ||
| Noise | Gaussian | ||
| Cost function | Mean squared error | ||
| Dropout | None | ||
Notes. The output of the encoder is first transformed into the barycenter of the anchor points using Eq. (14).
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.